An analytical model to prioritize a catalog of 17,000 medical tests at GDA

The GDA medical test catalog runs past 17,000 services. Publishing all of them is not a strategy, and publishing them alphabetically is not either.

At that size the important decision is not how each entry gets published: it is which ones get published first, because the order decides what people find for months afterwards.

And that decision was being made on judgment, not on data.

• Intuitive publishing: with no way to check which services have real demand.
• A scale no one can hand audit: 17,000 references are not reviewed one by one.
• Sensitive data: in healthcare, a badly loaded attribute is not a catalog error.
• Ground that shifted: showing up in a traditional search engine is no longer enough, you also have to be cited by generative engines.

A complete, badly prioritized catalog returns less than a partial, well ordered one.

We built an agent assisted analytical model that sets publication order by weighing real demand against clinical value, plus a cataloging pipeline that lets no attribute through unverified.

Continuous technical and semantic audit

• Ongoing diagnosis of the information architecture and the content.
• Aimed at both traditional search engines and generative engines, which cite rather than list.

The prioritization algorithm

• It weighs search volume, user intent, seasonality and service relevance.
• With that it defines which tests get published and in what order.
• It replaces intuitive publishing with a traceable decision: every service has a measurable reason to be where it is.

Competitive intelligence and a pyramid strategy

• Automated competitor benchmarking across the same universe of services.
• Content ranked into thematic clusters, so entries support each other instead of competing.

Cataloging with gates, because the data is clinical

• Agents for extraction, normalization and loading of technical attributes.
• Cross validation between agents and quality gates before an attribute reaches production.
• In healthcare the cost of wrong data is not a return: that is why the control sits before publishing, not after.

Architecture and technology

• A prioritization model over more than 17,000 references, fed by demand and intent signals.
• A multi-agent pipeline for extraction, normalization and curation with quality gates.

Learn more about artificial intelligence and software development.

The catalog stopped being published on judgment and started being published on evidence of demand.

What is running

• A universe of more than 17,000 services governed by a prioritization model.
• Publication order set by search volume, intent, seasonality and clinical relevance.
• Automated competitive benchmarking and content ranked into thematic clusters.
• A cataloging pipeline with cross validation and gates before production.
• Continuous technical and semantic auditing, aimed at traditional and generative engines.

What changes in the decision

• Every published service has a measurable reason to be where it is.
• Editorial effort concentrates where demand exists, instead of spreading evenly over 17,000 entries.

Why it matters

• In large catalogs the bottleneck is not producing content: it is deciding which to produce first, and that decision can be modeled.